Feature/fixing llm long prompt and model settings

[ayessipovskiyDAG]

feat:

• Handle null/undefined facet properties in mergeCountMaps to prevent crashes
• Pass prompts via stdin instead of command args to avoid length limits
• Make ANALYSIS_MODEL respect CODEX_INSIGHTS_MODEL env variable
• Remove unused imports
• Replace OPENAI_API_KEY hard gate with non-dry-run auth preflight
• Fail fast with clear errors when Codex CLI is missing or not logged in
• Keep --dry-run behavior unchanged (no auth check, no LLM calls)
• Document full-run vs dry-run auth behavior in README

[ayessipovskiyDAG]

I like this section 😂